1

Faculty of Engineering Technology

Route choice and speeds ofcyclists in Sao Paulo

Rutger MeesterBachelor Thesis - Final Version

20th of July 2018

Supervisors:Prof. Dr. Ing. K.T. (Karst) Geurs

Prof Dr. M.A. (Mariana) GiannottiDr. T. (Tiago) Fioreze

D.B. (Diego) Tomasiello MSc

Faculty of Engineering TechnologyUniversity of Twente

P.O. Box 2177500 AE Enschede

The Netherlands

Preface

In the curriculum of the Civil Engineering bachelor at the University of Twente, students com-plete a ten week research in which they use their gained knowledge of the past three yearsinto a project in the field of civil engineering. Personally, I have performed an internship atthe Escola Politecnica (POLI) of the University of Sao Paulo, which took place between Apriland July 2018.

Going to Sao Paulo gave me the opportunity to develop my skills in an environment whichwas completely new to me. Learning new things about the culture, language and peoplein Brazil, as well as working in a completely new atmosphere. Not only have the people inSao Paulo helped me finding my way with my thesis and keeping me in contact with otherparties, they also showed me how fascinating Sao Paulo and Brazil can be. Diego, Mariana,Tiago and Karst, a lot of appreciations for helping me to keep on track with my thesis andfinding explanations when I had doubts. Bruna and Leonardo, thanks for helping me exploreall the nice places and finding my way in Sao Paulo, making my stay way more enjoyable!

ii

Contents

1 Introduction 11.1 Project context . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.2 Objective . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21.3 Research questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21.4 Research method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

2 Theoretical framework 52.1 Cycling in Sao Paulo . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52.2 Factors of cycling speed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82.3 Factors of route choice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

3 Methodology 123.1 Data collection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123.2 Data processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143.3 Composing expected routes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

4 Data preparation 164.1 GPS point cleaning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164.2 Raw data quality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

5 Results 235.1 General characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245.2 Cycling speeds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265.3 Route choice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

6 Conclusion 32

7 Discussion 33

References 35

A Pedaladas app 38

B Survey questions 40

C Bike Network Sao Paulo 44

iii

List of Figures

1.1 Conceptual model research method . . . . . . . . . . . . . . . . . . . . . . . 3

2.1 Schematic overview of the cycling infrastructures in Sao Paulo . . . . . . . . 72.2 Spreading of cycling facilities among Sao Paulo . . . . . . . . . . . . . . . . . 72.3 Land use within the city of Sao Paulo . . . . . . . . . . . . . . . . . . . . . . . 102.4 Slopes within the city of Sao Paulo . . . . . . . . . . . . . . . . . . . . . . . . 11

3.1 Shared bicycle stations and their capacity . . . . . . . . . . . . . . . . . . . . 133.2 Method to process the raw data into the desired results . . . . . . . . . . . . 143.3 Visual overview of the route choice comparing . . . . . . . . . . . . . . . . . 15

4.1 Places of excluded points of trips . . . . . . . . . . . . . . . . . . . . . . . . . 184.2 Amount of excluded points per weather condition or temperature . . . . . . . 194.3 Amount of points per user . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194.4 Amount of captured and excluded points per day . . . . . . . . . . . . . . . . 204.5 Percentages of excluded points per reason per day . . . . . . . . . . . . . . . 204.6 Amount of excluded points per hour of the day . . . . . . . . . . . . . . . . . 214.7 Difference between collected and actual altitude of captured GPS points . . . 214.8 Difference between collected and actual speed between captured GPS points 22

5.1 Routes of users in Sao Paulo . . . . . . . . . . . . . . . . . . . . . . . . . . . 235.2 Amount of average trips and users per day . . . . . . . . . . . . . . . . . . . 245.3 Amount of average trips per weather condition . . . . . . . . . . . . . . . . . 255.4 Amount of average trips and users per day . . . . . . . . . . . . . . . . . . . 255.5 Speed of cyclists in Sao Paulo . . . . . . . . . . . . . . . . . . . . . . . . . . 265.6 Speed on different slopes in Sao Paulo . . . . . . . . . . . . . . . . . . . . . 275.7 Differences between lengths expected and collected route . . . . . . . . . . . 285.8 Differences between average percentage of distance on cycling infrastructure 295.9 Differences of distance on cycling infrastructure . . . . . . . . . . . . . . . . . 295.10 Difference between average amount of slopes on routes . . . . . . . . . . . . 305.11 Land use along the collected and expected routes . . . . . . . . . . . . . . . 31

7.1 Percentage of total cycling infrastructure per neighbourhood . . . . . . . . . . 33

C.1 Cyclable road network of the city of Sao Paulo . . . . . . . . . . . . . . . . . 44

iv

List of Tables

2.1 History of cycling in Sao Paulo . . . . . . . . . . . . . . . . . . . . . . . . . . 6

4.1 Points after data cleaning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164.2 Histogram of accuracy points . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

5.1 Speeds on road section in So Paulo . . . . . . . . . . . . . . . . . . . . . . . 265.2 Percentages for land use along routes . . . . . . . . . . . . . . . . . . . . . . 31

7.1 Automatic cycling trips detection . . . . . . . . . . . . . . . . . . . . . . . . . 34

A.1 Pedaladas options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

v

List of acronyms and abbreviations

BSS Bike Sharing SystemCEM Centro de Estudos da MetropoleGPS Global Positioning SystemNGO Non-Governmental OrganisationPOLI Escola PolitecnicaRMSP Regiao Metropolitana de Sao PauloUSP Universidade de Sao Paulo

vi

Abstract

Over the past decades, cycling has become a more important aspect of traffic in Sao Paulo.In order to increase the number of bike users in the city, different improvements have beenintroduced. New infrastructure has been built to guarantee cyclists their own spot on theroad, new regulations have been implemented to guarantee safety and awareness to otherroad users, as well as more room for cycling on Sundays and holidays to introduce cyclingto new users.

This thesis focuses on the collection, processing and analysing of cyclist trips in the city ofSao Paulo. Goal of the thesis is to gain more insight in the different aspects of cycling speedand route choice within Sao Paulo. The approach of cycling parties CicloCidade, BikeAnjoand Tembici, along with inviting people directly from the streets, led to 26 volunteers in-stalling the Footprints travel app. Only 6 people made a total of 59 useful trips that could beused within this research. Almost 40% of the GPS points was not useful for data processing.

A GIS comparison between the expected and collected routes showed a distance favour of15% for collected routes. 75% of the road networks contained an average speed between 10and 30 km/h. Highest speeds are being obtained on flat surfaces, with speeds decreasingas slope increases both uphill and downhill.

Expected and collected distance on cycling facilities had a large overlap, with both routeshaving on average respectively around 10% and 9% of their distance on a cycling lane (ci-clofaixa). Expected routes (14%) had, compared to the collected routes (10%), on averagemore distance on a cycling boulevard (ciclovia). Most common land use along the routesis residential, however when taking into account the average land use in Sao Paulo cyclistsprefer to cycle along mixed, commercial and public land use. Cyclists take on average aroute that is more hilly than (35% of routes), or as hilly as (25% of routes) the expectedroute.

The external annotation environment and following survey have only been used by a minorityof the volunteers and the data from these sources is therefore not used. Future researchcan contribute to the best way of approaching and making volunteers use the app, to obtaina larger and more scattered sample. As well as separating car and bicycle trips better orfinding solutions to increase cycling experiences on specific route sections.

vii

Chapter 1

Introduction

This thesis presents an approach to the collection, mapping, and analysis of cyclists expe-riences in Sao Paulo, starting with the associated theoretical framework. This research willexplicitly focus on cycling speeds and cyclists’ route choice within the municipality of SaoPaulo.

1.1 Project context

Disparity is a very common appearance within the city of Sao Paulo. According to Watts &Forner (2017), at least 15.000 people are living on the streets. On the other hand Sao Paulohas the largest helicopter fleet in the world, indicating the amount of wealthy people living inthe city. This disparity does not benefit the daily living in the city, and diverges the inhabitantsfurther away from each other. To gain more equity among the inhabitants, everyone shouldbe given a minimal solution to their needs. One of the basic and most important aspects ofequity will be the one regarding accessibility. Despite the transport mode that the inhabitantuses, all people should be given opportunities to travel in a comfortable and safe way. In or-der to measure this equity among the inhabitants, more knowledge should be gained aboutthe different transport modes. Especially more information is needed about cycling, sincethere is not much information available on cycling within the city.

To gain more knowledge about cycling in Sao Paulo, a GIS layer with a cycling network hasalready been made as a result of a collaboration between the universities of Sao Paulo andTwente (Pritchard 2018). In this network, travel distances and times for different routes canbe generated, by taking into account the network density, hilliness of the network, availabilityof dedicated biking infrastructure and the delay at intersections. Thus, a working model withaverage to close predictions of the network speeds in the city is present. However, the modelonly takes estimated constraints that are present in the network into account. Used data inthe model relied on Dutch cycle speeds, whereas these speeds are very different from theones in Sao Paulo. Therefore further research is needed to come up with the parametersthat influence route choice and therefore network speeds.

1

1.2 Objective

The main objective of this research is to gain more knowledge about the characteristics ofcycling trips in Sao Paulo. Looking at the fact that there are many uncertainties about thecurrent travel times in Sao Paulo, more knowledge is needed to get a better calibration of theconstructed model (Pritchard 2018). How can the different trips in Sao Paulo be captured,what characteristics of the network influence the cycling speed and what causes the cycliststo take certain routes? Answers on these questions will give more insight on the travel timesof cyclists on the network and therefore a better working model.

1.3 Research questions

In order to reach the objective stated above, a main goal and some different sub goal arecomposed. These goals will give directions to the research, as well as some clear goals thatshould be reached. Methods to reach these research aims can be found at the end of thissection in Figure 1.1.

Main research aim of this thesis is:

‘To gain more knowledge about the factors that influence the route choiceand cycling speed of cyclists in Sao Paulo’

This main aim can be divided into different sub aims. First of all, more knowledge aboutroute choice in general has to be gained. Specifically, route choice in urban areas shouldbe investigated, since this research will focus on the city of Sao Paulo. With this informationthe different characteristics to look at can be composed.

‘What factors influence cyclists’ route choice in urban areas?’

Additionally, the same knowledge should be gained about the factors that cause longertravel times in urban areas and Sao Paulo specifically. Therefore more insight in the cyclingspeeds in Sao Paulo will be obtained, as well as the characteristics that influence thesecycling speeds.

‘What factors influence cycling speeds in urban areas?’

Also, since this research will mostly work with GPS data, more information about the col-lected points is necessary. When is the GPS data useful for processing, what informationcan be gathered when using the GPS data and is all information that is being collectedreliable enough?

‘What data can be collected with the collection of GPS points, and what are theconstraints when working with this data?’

A very important sub-question is the one about route choice and cycling speeds in Sao Pauloitself. Determining the commonly used routes is key in finding the decision why cyclists takethis routes, as well as finding the network speed on the chosen routes.

2

‘Which routes do cyclists take in Sao Paulo, and what is the cycling speed onthese routes?’

Additionally, after finding the preferred routes in Sao Paulo, finding the cause of havingcertain route preferences is necessary. Why are certain routes more popular than others andwhy do these characteristics have influence on the route choice. Especially the influence ofcycling speeds on route choice should be investigated.

‘What is the motivation on routes chosen by cyclists in Sao Paulo?’

1.4 Research method

To get the answers on all the research questions, a conceptual model has been made thathelps giving the research directions. This model can be found in Figure 1.1 and is furtherclarified in the text underneath.

Figure 1.1: Conceptual model research method

The first step in the process is to find more information on the cycleability in urban areas.This will be done by conducting a literature study on research that has been performed inother metropolitan areas. Besides the characteristics, also more information about the sta-tus of cycling in Sao Paulo should be gathered. This will help in evaluating the current habitsof cycling trips in Sao Paulo, and will give a clear overview of what role cyclists play in traffic.

Furthermore, the biking infrastructure in the city of Sao Paulo needs to be obtained to findout what possible routes can be taken. Besides the biking infrastructure, environmental char-acteristics about the city of Sao Paulo should be gathered. The characteristics that shouldbe taken into account are the ones that have any impact on cycling speeds and route choicewithin the city. Characteristics that will be useful for this thesis will be explained in Chapter 2.

3

Additionally, trip data will be collected with a travel app. This app collects GPS points withthe use of their own mobile device. The app will help in finding the routes taken within SaoPaulo, and how they deviate from the expected routes. More information about the traveldiary app can be found in Section 3.1.

With the use of the app two different routes can be composed. One route with just the firstand the last captured GPS points which represents the expected route, the route which hasbeen composed with the use of the bike network of Sao Paulo (Pritchard 2018). And anotherroute with the use of all the collected GPS points of the trip, which will represent the actualroute the cyclists has taken.

After collecting sufficient data with the app, the cycling speed on the routes of the cyclistscan be measured. Also a comparison between the expected and the collected route can bemade. Within this comparison differences in chosen routes will be searched for, along withdifference in the characteristics that identify these routes.

4

Chapter 2

Theoretical framework

Within this section, earlier research is being used to gain a better foundation for this re-search. First of all the current status of cycling in Sao Paulo has been investigated. Secondlythe characteristics that influence the cycling speed are researched and finally the things thatinfluence route choice are examined.

2.1 Cycling in Sao Paulo

The streets in Sao Paulo have always been characterised by a distinct hierarchy betweenmotorised and non-motorised vehicles. Having to share the network with the motorised traf-fic, cyclists are mostly expelled to the roadsides with many obstacles like lampposts, waterdrainage systems or parked cars. Also, on the side walk on the other hand, cyclists con-flict with pedestrians and suffer from the kiosks and streets vendors that are present. Thus,apart from the specifically designated biking infrastructure that is only present in a minorityof the streets, cyclists often do not have a reserved spot on the streets. Cycling thereforecannot be perceived as safe within Sao Paulo by a number of cyclists.

Not only perceived safety, also actual safety is a concern. Within the city of Sao Paulo, cy-cling trips represent less than 1% of the total daily trips (Sa et al. 2016). However, cyclistsfatalities in road accidents are much higher at 7% of total victims, pedestrians not included(Lemas & Neto 2014). This suggests that cyclists lack a lot of safety on their routes. Safetyhowever is rising within Sao Paulo, with cycling deaths decreasing from 0,76 to 0,56 per1000 inhabitants between 2007 and 2012 while having a rise in the amount of total distancetravelled by bike (Sa et al. 2016). Not only cycling safety is an issue in Sao Paulo, bike theftand robberies also take place more than average in Brazil. According to Freitas & Maciel(2017), 43.6% of the respondents in their survey on cycling in Brazil declared that they hada bicycle stolen.

Another issue within the city of Sao Paulo is the perceived air quality. Images of skyscrapersconcealed in clouds of smog, as well as large piles of garbage on the sides of the street orin rivers are not unusual in Sao Paulo and restrain the inhabitants from seeing cycling as

5

healthy. Nevertheless, de Hartog et al. (2010) mentions that exercising by means of cyclingoutweighs the risk caused by air pollution. This reference also states that cycling exercise ona daily basis would benefit the life expectancy by 3 to 14 months, while inhaling pollutantswould harm life expectancy by only 1 to 40 days. Also, air quality has improved over thefew decades in Sao Paulo. As stated by Andrade et al. (2017) most of the concentrationsof primary pollutants have decreased over the last 30 years. The main issue right now isto decrease the emissions of ozone and fine particles which are mostly there because ofthe transportation sector. Ribeiro et al. (2015) complements these statements by inform-ing about the fact that ozone particles are currently the most frequently exceeded emissionstandards.

As declared by Rolnik & Klintowitz (2011), average car speed in Sao Paulo during morningand evening peaks is respectively 27 and 22 km/h. This indicates that cyclist can com-pete with driving motorised vehicles during working hours. Governmental plans as well ascyclist groups, caused a lot of new options to start cycling within Sao Paulo over the pastdecades. Mainly, three different phases can be distinguished (Lemas & Neto 2014). First ofall the phase where first plans and projects of cycling infrastructure were presented. Secondphase indicated many legal concerns on cycling, such as defining bicycles as vehicles. Inthe last phase a large increase of cycling infrastructure construction started. A schematicoverview of these phases can be found in Table 2.1.

Table 2.1: History of cycling in Sao Paulo (Lemas & Neto 2014)

Phase 1:Early 1980to early1990

• First municipal plan indicating a sectoral cycling infrastructure system.• First project for cycling infrastructure.• Cycling is strongly influenced by the newly published Brazilian manual

for planning cycling infrastructure.Phase 2:Early 1990to late2000

• First legal instruments concerning cycling infrastructure, first bicyclepaths and the first version of the Critical Mass protest; demandingmore rights for cyclists.

• New Brazilian Traffic Code, defining bicycles as vehicles andguaranteeing cyclists the rights to ride on streets and circulate intraffic.

• Three new municipal laws: Obligated space for bike paths on newroad infrastructure; Bike paths must be part of the road system;Indication of a cycling network to be implemented.

Phase 3:Late 2000to current

• Remarkable increase on construction of cycling infrastructure, yet stillunconnected.

• Foundation of the two major commuting-cycling advocacyorganisations.

• Operational Bike Lanes are being created, leisure space on mainroads on Sundays.

6

The amount of cycling trips in Sao Paulo has almost doubled between 1997 and 2007(Lemas & Neto 2014). Mostly constructed between 2012 and 2017, different kinds of cy-cling infrastructure are present within Sao Paulo. For a schematic overview, see Figure 2.1.The spread of the infrastructure among Sao Paulo can be found in Figure 2.2. As seen inthis Figure, most of the cycling infrastructure in Sao Paulo is located near the city centre andon the marginal and radial routes, the main access roads to Sao Paulo.

Figure 2.1: Schematic overview of the cycling infrastructures in Sao Paulo(Orenstein 2017)

Figure 2.2: Spreading of cycling facilities among Sao Paulo (Centro de Estudos da Metropole 2018)

7

Sao Paulo’s biking infrastructure is centered around the cycling paths, ciclovias, which arepresent in some of the cities districts. These ciclovias are only accessible by bike and arein most cases physically separated from other traffic. However, these lanes are mostly con-structed on the central reservation, after the original driveway was finished. Therefore it hasto avoid obstacles like trees and lampposts, causing it to have many turns instead of goingin a straight line. Another downside of the ciclovia is that they are often very narrow, makingit impossible to overtake low-speed cyclists or pedestrians, who often use ciclovias as well.

Another cycling facility within the city of Sao Paulo is the ciclofaixa. This facility can be com-pared with a cycling lane, which is only a marked area on a street designated for cyclists.Cyclofaixas are always combined on one side of the street, causing users to have the needto cross the street to get on one. The ciclofaixa often has no physical separation from themotorised vehicles however it is sometimes being separated with cat’s eyes or turtles, alarger variant of the cat’s eyes.

The last and most uncommon cycling infrastructure is the ciclorota. This cycling facility hasno physical characteristics whatsoever, however painted icons or signs are present that in-dicate that these facilities are commonly used by cyclists. Also these streets often havetraffic calming facilities, such as narrowed roads and speed bumps, in combination with alow maximum allowed speed.

On Sundays and holidays cyclists are able to use even more streets with the use of socalled ”conviva” streets, meaning live together in Portuguese. These cycling facilities makeup almost 30% of the total cycling infrastructure (Lemas & Neto 2014), causing a huge liftin availability. During these days one driveway on the main streets is turned into a cyclinglane, offering safety with the use of pylons and traffic controllers. These days are often usedby inexperienced cyclists to learn to cycle, as well as by experienced cyclists for leisure tripsaround the city.

2.2 Factors of cycling speed

Many factors on the route are influencing cycling speeds and therefore travel time on thenetwork. In this section, the different factors defined in the literature are discussed.

Junctions: As described by Menghini et al. (2010), the amount of stops because of junctionsis increasing the total travel time on the cycling network. It also mentions that especially theamount of signal controlled junctions have negative impact on the speeds of cyclists. This isalso characterised in Buehler & Dill (2016), which on contrary also states that bicycle spe-cific traffic devices such as bike boxes, bike traffic signals and bicycle signal activation havea positive influence on the cycling speeds at junctions.

8

Traffic volumes: Both Broach et al. (2012) and Sener et al. (2009) consider traffic volume tohave a large impact on cycling speeds. Especially on non-separated bike lanes the influenceof the mixed traffic volume was significant, as a result of less space on the cycling facilities.The researches also show that high speed limits on the adjacent road have a large negativeimpact on the cycling experiences, however only if the cyclists are not experienced in bikingand have trips of five miles or shorter. Not only traffic volume but also pedestrians, who alsomake use of the cycling infrastructure, have an influence on the network speeds of cyclists.Regarding to Bernardi & Rupi (2015), a drop in cycling speed of more than 10% is presentwhen a pedestrian is encountered on the route.

Cycling facilities: Many studies have proven that the use of separated cycling infrastruc-ture improves the cycle speed on the network (Broach et al. 2012) (Buehler & Dill 2016)(Heinen et al. 2010) (Hood et al. 2011) (Pucher & Buehler 2008) (Zimmerman et al. 2016).Besides,(Dill 2009) states that in most of the trips separated bicycle facilities were beingused in their study, implying that these facilities provide fastest connections. Sener et al.(2009) also describes that small lane widths and the absence of bicycle lanes cause longertravel times on both short and long commute distances.

Pavement conditions: There are not many studies that identify the role of the quality ofthe facility. Mostly because the fact that the quality of the pavement is location specific andmight vary within a road section. However, Brezina et al. (2012) mentions that differencesbetween road surfaces can lead to an increase in energy expenditure of more than 100%.From this we can conclude that cycling speeds will deviate on different road surfaces, withpreferences for good lane evenness and plain asphalt.

2.3 Factors of route choice

Some different factors can influence the cycling experience in a positive or negative way. Asstated by Kuiper & Braakman (2009) and Sustrans (2014) five different requirements can besummarized to have a look into the wishes and demands of cyclists.

Attractiveness: Infrastructure has to fit in the surrounding view and atmosphere, in order tomake cycling carefree and pleasant. However, peoples view on attractiveness is determinedby a wide range of factors, which may be more or less important to individuals. Bernardiet al. (2017) states that most trips are being made on links with good to very good level ofbeauty. On the other hand Snizek et al. (2013) ensures that the amount of natural aspectsadjacent to the cycling infrastructure, influences the attractiveness of the network in a posi-tive way. (Chandra et al. 2017) poses that the amount of criminal activity has a large impacton the attractiveness of the infrastructure, causing cyclists to avoid areas with high recordsof criminal activity. An overview of the land use within Sao Paulo can be seen in Figure 2.3.

9

Figure 2.3: Land use within the city of Sao Paulo (GeoSampa 2018)

Coherence: The cycling infrastructure has to provide connections between origins and des-tination, such that the network forms a coherent whole. As described by Winters et al. (2013)the amount of intersections also influences the coherence of a network, since highly con-nected road networks allow more routes and efficient travel. Looking at earlier statementsabout junctions, we can conclude that they are great for different route choices but less opti-mal looking at cycling speeds. Also Dill (2009) states that cycling is a main mode of transportin well connected street grids.

Comfort: Nuisance and delay caused by bottlenecks, crowded routes and shortcomings inthe infrastructure do have a negative impact on the cyclists’ travel experience. Additionalphysical effort produced by this delay discourages cycling. Besides, the terrain can have anenormous impact on the cyclists’ experience on the biking network. As seen in Figure 2.4,the amount of slopes in Sao Paulo is critically high. This will cause a large fluctuation incycling speed depending on the direction, as well as a large loss of human energy. Howeverin the figure can also be seen that most of the cycling infrastructure within Sao Paulo hasbeen build on places with the least amount of slopes.

Directness: Perhaps the most important factor of the chosen route is the distance that hasto be covered. Cyclists need to be offered the most direct route, therefore keeping detoursto a minimum. Dill (2009), notes that the directness is seen as an indicator for how complex

10

Figure 2.4: Slopes within the city of Sao Paulo (Trimble Data 2015) (Centro de Estudos da Metropole 2018)

the route is and thereby indicates whether the cyclist had to make choices between left orright several times. The necessity to make left or right turns suggest that the directness ofthe route is diminished.

Safety: Cycling infrastructure should guarantee the safety of cyclists and other road usersin traffic. Since cyclists are relatively vulnerable, road design has to take this factor intoaccount. As stated in Broach et al. (2012) and Fishman (2016) cyclists, women especially,have a preference for separated bike paths and bicycle boulevards. Striped bike lanes areonly preferred when low-traffic neighbourhood streets are not an option, however they arehighly valued compared to streets without the striped bike lanes. Lowry et al. (2013) statesthat also the width of the outside lanes, through lanes and the shoulder, vehicle parkingalong the biking facility, traffic volume, traffic speeds, amount of heavy vehicles on the out-side lane and the height of the curbs influence the safety of cycling on mixed traffic lanes.

11

Chapter 3

Methodology

Within this chapter the main principles of the data collection and processing will be ex-plained. Starting with an explanation of the different data that will be collected, followed bythe processing of the data.

3.1 Data collection

To gain the most knowledge about cycling in Sao Paulo, as much volunteers as possible havebeen asked to collect data for the research. Three different parties have been approached,which are further defined in the itemisation below. To grow the amount of volunteers evenmore, flyers have been handed out to cyclists in the Faria Lima neighbourhood to invite themto help collecting data. Faria Lima is a location with a lot of different cycling purposes, suchas commuting cyclists that use the metro and transfer to a bike or leisure cyclists that usethe excellent cycling infrastructure in Faria Lima.

BikeAnjo: A group of biking fanatics founded in 2010 that helps new people to ride a bike,as well as organising cycling trips around the city of Sao Paulo. With the use of BikeAnjo alarge variety of cyclists can be reached, since both experienced and inexperienced cyclistswill be part of this community. It is expected that deviations of the mean speeds will be high,but the average speed tends to be a good representation since there will be cyclists fromboth sides of the spectrum. However, as stated by Stinson & Bhat (2004), there is a differ-ence in route choice factors between different levels of experience. Inexperienced cycliststend to give more value to factor related to separation from cars, while experienced cyclistsare more sensitive to factors related to travel time.

CicloCidade: A non-profit organisation that focuses on constructing and contributing to amore sustainable city. Its aim is to help cyclists provide comfort and safety on their trips.Volunteers from CicloCidade will most likely be experienced cyclists that make trips for theirdaily exercising and for leisure. The expectation is that speeds of these cyclists will be higherthan average, since these persons are used to cycling in Sao Paulo and will know how toreact in specific situations.

12

Tembici: The main operator of the BikeSampa bicycle shared system (BSS) in Sao Paulo.With biking stations mostly in the centre and south western part of the city, trips will mainlybe generated in these areas. Places and capacity of the docking stations of Tembici canbe found in Figure 3.1. As seen in this figure most of the large stations are near the publictransport station, where commuters take the bicycle for the distance to their job or school.Trips made with the Tembici bikes will most likely be leisure trips or commuters that facilitatethe first or last mile to public transport stations.

As mentioned in Bikeplus (2017), nearly 25% of the people that use the BSS in their surveystarted or restarted cycling with a break of at least 5 years, indicating the amount of inex-perienced cyclists. Also bus and train implementation in their trips was very common, withrespectively 25 and 40% of the respondents stating to use the bike in conjunction with thebus or train. Almost 70% of the shared cycling trips were between 2 and 4 miles long, withrespectively 21 and 18% being made for commuting or for leisure or exercising.

Figure 3.1: Shared bicycle stations and their capacity

With the use of a tracking app (MEILI Devs 2018) it is possible to capture the routes of thevolunteers, with the use of a GPS technology in their mobile devices. The GPS points makesit possible to determine the travelled route of the users. Also an annotation process is im-plemented, making the user able to correct and comment on the made trips. With the use ofthis annotation environment more information about their experiences on the road, purpose,destinations and transport mode can be collected. A complete overview of the aspects ofthe trip that can be annotated can be found in Appendix A.

13

To gain even more knowledge about the perception on cycling in Sao Paulo, a follow-upsurvey has been held among the volunteers. Within this survey their stated preferenceon route choice and cycling speeds, as well as socio-economic characteristics have beencollected. With the use of the data of the survey, which has been linked to the people’strips, more information can be gathered about the different volunteers and therefore theinhabitants of Sao Paulo. The full survey set-up can be found in Appendix B.

3.2 Data processing

Data processing is necessary to come to the desired results in a later stadium of the re-search. To do so, a flow chart has been produced in Figure 3.2 that shows the processingof the collected data in a schematic way. This scheme shows the data cleaning of the GPSpoints, as well as the process from point to trip to results for every trip the cyclists have made.

As seen in this figure, first thing to do is to exclude all invalid point. This which will befurther clarified in Section 4.1. Secondly every trip that has been captured by the app willbe selected. After selecting the trip there are two possibilities. Depending on the calculationbeing an expected route or a collected route either all of the points or just the first and lastpoint that intersect with this line will be selected. These points are being used to solve theroute with the use of the network, as well as to calculate the slope and speed between thesepoints. The route that has been composed will then be used to measure the overlap withcycling facilities and the neighbouring land use to find statistics on these characteristics.

Figure 3.2: Method to process the raw data into the desired results

14

3.3 Composing expected routes

The process from GPS points to routes via a model, further called the expected routes, isprobably best explained by Frejinger (2008). A visual description of this process can be seenin Figure 3.3.

Figure 3.3: Visual overview of the route choice comparing (Frejinger 2008)

Starting with the GPS points and the cycling network of Sao Paulo, different paths can bemade between the points. For this research only the first and last captured points of a tripwill be used for the processing of the expected route.

These points have been implemented in a dense and complete biking network layer of SaoPaulo, which is available within the GeoLab of the University of Sao Paulo. A full overview ofthis network which is used in the data processing can be seen in Appendix C. The networkhas been created by the ”Centro de Estudos da Metropole” (CEM), located in Sao Paulo.CEM is an institution for advanced research in social sciences and investigates themes onthe subject of inequalities and the formulation of public policies in contemporary metropolises(Centro de Estudos da Metropole 2018).

For the route composing a link elimination approach will be used. This approach consistsof the shortest distance path with respect to different attributes of the route. Ben-Akiva &Bierlaire (1999) mentions that not every attribute in the choice set necessarily has to be adirectly measurable one, and that attributes have to be tested before finding the appropriateattributes to work with. The trip will be generated by inserting only the beginning and endingpoints of the collected trips in the network. Attributes of the route that are taken into accountin the approach for Sao Paulo are presence of dedicated biking infrastructure, slope, trafficvolume and the waiting time at intersections Pritchard (2018).

15

Chapter 4

Data preparation

This chapter focusses on the processing of the raw data points. What points can be excludedfrom the processing, based on a number of rules that are applied the GPS points. What isthe quality of the data captured by the app and what can already be concluded from thisdata.

4.1 GPS point cleaning

Not all GPS points will be useful to use within the data processing. Table 4.1 shows theamount of points that have been excluded from the data, as well as the reason for theexclusion. Points can be excluded twice for different reasons, causing the total amount ofexcluded points to not match the sum of the different excluded points. Underneath the tablemore information about the rules is given.

Table 4.1: Points after data cleaning

Description of rule Amount of points Percentage

Points in raw data 16912 100%Points not in Sao Paulo 1242 7,3%Points not assigned to a trip 3030 17,9%Points with accuracy less than 20m 1442 8,5%Points within trip with speed higher than 50 km/h 2504 14,8%Points useful for data processing 10478 62,0%

Points not in Sao Paulo: All points that do not intersect with the shapefile of the munici-pality of Sao Paulo have been excluded from the database. Trips that start within, but endoutside the municipality boundaries, are therefore used within the data processing until thetrip leaves the city borders.

Points not assigned to a trip: All points that do not intersect with a trip as they appearfor commenting in the annotation environment have been excluded from the database. Tworeasons can be the cause for this. Either the algorithm within the app does not include all

16

points to a trip, since it judges the point to be wrongly captured. Or the algorithm does notcapture a whole trip, however the reason for this could not be found.

Points with accuracy less than 20 meter: After looking at the histogram of the accuracy ofthe points, which can be seen in Table 4.2, it can be concluded that when taking a boundaryof 20 meter accuracy the sample will still include 90% of the total points.

Table 4.2: Histogram of accuracy points

Bin Frequency Percentage5 3 0%10 8783 52%15 2819 17%20 3723 22%30 728 4%40 258 2%50 114 1%75 149 1%100 124 1%More 210 1%

Points within trip with speed higher than 50 km/h: If a point within a trip reaches a speedhigher than 50 km/h, it can be concluded that this trip has been made by a motorised vehicle.Therefore the whole trip will be excluded within the data processing if a point within this tripreaches a speed of more than 50 km/h.

17

4.2 Raw data quality

Within this section, more information about the excluded points will be given. Differences inplace, time, conditions will be searched for, as well as differences between the users.

Excluded points per place: Different points have different reasons to be excluded. Figure4.1 shows the places of these points. As seen in the figure most of the excluded points be-cause of the high speed are on two roads, which represent major roads in Sao Paulo. Alsono specific places in the city could be found that have clusters of points with low accuracy.

Figure 4.1: Places of excluded points of trips

Excluded points per weather condition: Figure 4.2 describes the average amount of col-lected and excluded points per day for different weather conditions. Days are being split upinto days with clouds, spells or predominant sun, as well as into days with different averagetemperatures. The most amount of collected points, as well as the least amount of excludedpoints can be found on days with sunny periods. On cloudy days the highest percentage ofpoints have been excluded. On days with an average temperature of higher than 21 degreesthe least amount of points were being collected. However the percentage of excluded pointsper day was the highest on days with an average temperature of less than 17 degrees. Thisgoes in line with the percentage of excluded points in cloudy days, since on a cloudy day thetemperature will also be low.

18

Figure 4.2: Amount of excluded points per weather condition or temperature

Excluded points per user: From the 26 people that installed the app, only nine peopleactually collected points with their device. As seen in Figure 4.3, three of these nine usersdid not collect enough points to compose trips, leaving only six people with useful trips to beused. No link can be made between the amount of points collected and the percentage ofuseful points.

Figure 4.3: Amount of points per user

Excluded points per day: Figure 4.4 shows the amount of collected and excluded pointsper day. On average, most points have been collected on Thursday and Friday. Also seen in

19

this figure is that points collected on Friday or Sunday will most likely be excluded from thedatabase. The reasons of this exclusions can be seen in Figure 4.5, which shows that onSunday most of the points have been excluded because of non-cycling trips while on Fridaya mix of reasons causes the points to be excluded.

Figure 4.4: Amount of captured and excluded points per day

Figure 4.5: Percentages of excluded points per reason per day

20

Excluded points per hour: Figure 4.6 illustrates the amount of collected and excludedpoints per hour of the day. As seen in this figure the most amount of points have beencollected between 10h and 12h in the morning, as well as between 18h and 21h. It can alsobe concluded that the least amount of points have been excluded in the early morning, aswell as on hours where many points have been collected.

Figure 4.6: Amount of excluded points per hour of the day

Captured altitude: The footprints app collects an altitude via the mobile device of the vol-unteer. To get more information about the quality of the data, the captured altitude has beencompared with the altitude of the elevation map of Sao Paulo (Trimble Data 2015), which isbeing assumed to be correct. As seen in Figure 4.7 the two altitudes do differ. Collectedaltitudes are on average almost ten meters lower than the ones considered as correct.

Figure 4.7: Difference between collected and actual altitude of captured GPS points

21

Captured speed: The Footprints app collects a speed between two collected points. Toget more information about the quality of this data, the actual speed has been calculated byfinding the distance between two points using the longitude and latitude coordinates, dividedby the difference in capture time of the GPS points. Figure 4.8 shows the difference betweenthe collected and the calculated speed, showing again that the collected speed is lower thanthe calculated one. In the figure it can be seen that all high speed points have been excludedfrom the data processing. Also a large amount of collected points with a speed of zero km/hcan be seen, showing the inconsistency of the speed collecting device.

Figure 4.8: Difference between collected and actual speed between captured GPS points

22

Chapter 5

Results

During the data collection, a total of 26 volunteers have installed the app on their device.However only 9 volunteers made trips that could be used within this research. A total of59 trips could be used within this thesis, which can be seen in Figure 5.1. Almost none ofthe volunteers made use of the annotation environment which was made available on theinternet, therefore this data is not used. Because the survey has only been filled in by 6of the 26 volunteers, results of the survey have not been included in this section. At firstthe general characteristics of the trips have been composed, secondly the speeds will bediscussed and at last the route choice of the cyclists will be evaluated.

Figure 5.1: Routes of users in Sao Paulo

23

5.1 General characteristics

Within this section the cyclists’ trips have been evaluated. Days and weather condition ofthe captured trips, amount of data per user and time of trip start have been distinguished.

Figure 5.2 shows the amount of users and trips per day or kind of day. As seen in this figuremost trips are being made on days early in the week, while less trips are being made onFridays and on weekend days. Especially Sunday is a day with almost no captured trips. Aconclusion for this could be that most of the trips have a commuting purpose. On averagethe amount of users per day is more or less the same, except for Sundays. Thus the highamount of trips on certain days is being caused by only a few cyclists that make differenttrips.

In Figure 5.3 the amount of trips per weather condition can be found. This figure shows thatmost of the trips are being made on days with mediocre weather conditions, days that aremostly sunny and have average temperatures between 17 ◦C and 21 ◦C.

Figure 5.2: Amount of average trips and users per day

24

Figure 5.3: Amount of average trips per weather condition

The amount of starting trips per hour can be found in Figure 5.4. This figure shows that mostof the trips are being started during daytime, as well as an avoidance of the morning rushhours. Cyclists start a lot of trips around six o’clock in the afternoon, probably because thisis close to sunset and close to ending times of workdays as well.

Figure 5.4: Amount of average trips and users per day

25

5.2 Cycling speeds

The data from the collected GPS points has been used to determine the speeds on thenetwork. With the longitude and latitude coordinates, the distance between points has beenfound. Also the duration between two point collections resulted in the time spent. Both thedistance and time have been used to compose the speeds for every road section of thenetwork. Figure 5.5 shows the different average speeds of the road sections in Sao Paulo.As seen in this figure, there is a large difference between the different routes in Sao Paulo.Especially in the city centre, presented by the north-eastern trips, the average speed on roadsection is low. Parts in the south-western area of the research area show higher speeds.Table 5.1 shows the spreading of the speeds, which presents that 75% of the road sectionshave a speed between 10 and 30 km/h.

Table 5.1: Speeds on road section in So Paulo

Speed Amount of road sections PercentageBetween 0 and 10 km/h 1461 14,1Between 10 and 15 km/h 2050 19,8Between 15 and 20 km/h 2645 25,5Between 20 and 30 km/h 3315 32,0More than 30 km/h 904 8,8

Figure 5.5: Speed of cyclists in Sao Paulo

26

Speeds on slopes

Now that average speeds on road sections are found, the different speeds on distinct char-acteristics can be evaluated. Since the slope of the road section will most likely have a largeimpact on the speed, the relation between these two characteristics has been examined.Figure 5.6 shows the relationship between the speeds and the different slopes between -20and 20 percent.

This figure shows that speed will decrease as the slope uphill increases. The fact thatspeeds decrease when cycling uphill on a larger slope can be explained rationally. Afterall, an increase of slope will require more effort from the cyclist. Largest speeds are beingobtained on slopes downhill varying from 2 to 6%, while higher slopes downhill will cause adecrease in speed. This can be caused by the cyclists feeling the urge to break more if theslope is increasing, causing the cyclists to have a lower speed. Another hypothesis can bethat higher speeds on nearly flat surfaces are not being obtained on these roads becauseof the slope but because of the type of infrastructure. The main cycling infrastructures havebeen created mostly on flat to nearly flat surfaces and can benefit from a better traffic flow,more green-time on junctions and a higher road capacity, which will cause an increase incycling speed.

Figure 5.6: Speed on different slopes in Sao Paulo

27

5.3 Route choice

Within the theoretical framework a lot of different characteristics that influence route choicehave been found. In this section four of these characteristics will be evaluated for the trips inSao Paulo: Total distance, use of the different cycling facilities, slope of the route and landuse along the road. Where possible, a 95% confidence interval has been calculated andadded to the figures.

Distance

As mentioned earlier the characteristic of the route that has the most impact on the chosenroute is the distance. Using either all the captured GPS points or only the starting and endingpoint of a trip the different routes have been composed. Figure 5.7 shows the differencebetween the average length of the collected and the expected route. As seen, the averagelength of the collected trips is nearly 15% higher than the expected trip. Thus people arelikely increasing their travelled distance in order for them to have other benefits on theirroute. Another possible explanation is the fact that many cyclists in Sao Paulo use the bikefor leisure or exercising trips.

Figure 5.7: Differences between lengths expected and collected route

28

Cycling facilities

The overlap between both the routes and the cycling facilities has been measured to calcu-late the differences between the collected and the expected route. The average numbers forthe routes on the cycling facilities can be found in Figure 5.8. App users make on averageless use of ciclovias and ciclofaixas on their routes, however the amount of distance on aciclovia differs more. Figure 5.9 shows the difference between the distances on cycling fa-cilities per route.

Figure 5.8: Differences between average percentage of distance on cycling infrastructure

Figure 5.9: Differences of distance on cycling infrastructure

29

Slope

Large slopes will require a lot more effort from the cyclists. Cyclists may be avoiding theseslopes by composing their routes via other areas. Figure 5.10 displays the difference ofslopes between the collected and expected routes of the cyclists. This figure indicates thatthere is a large difference between the collected and expected routes, with expected routesbeing on average less hilly. Thus cyclists in Sao Paulo possibly do not take steepness intoaccount as much as the model expected. These indifferences might also be caused bydifference between the slope calculations of the two different routes. Slopes of collectedroutes have been composed using the altitude of the GPS points. Expected routes didnot include the GPS points and therefore the elevation raster map of Sao Paulo has beenused to determine slopes on these routes. As stated in Section 4.2 there are differences inelevations between these two methods which may cause this variation.

Figure 5.10: Difference between average amount of slopes on routes

Land use

With the use of a buffer around the composed routes the existing land use along the routeshas been found in a 50 meter radius, so that information of the areas immediately alongthe streets could be found. Figure 5.11 and Table 5.2 show the percentages of the averageland use of both routes, as well as the average land use in the whole city of Sao Paulo. Asseen in the figure there is not much difference between the collected and the expected route.The land use along the routes is mainly residential area, however the residential area is stillhigher for Sao Paulo on average. Percentages of land use along the route that are higherthan average for the routes are the commercial, public and mixed land use.

30

Figure 5.11: Land use along the collected and expected routes

Table 5.2: Percentages for land use along routes

Group Land use Collected route Expected route Sao Paulo average

No informationNo information

Other1,2 1,6 8,5

ResidentialResidential high-riseResidential low-rise

46,9 45,2 57,0

CommercialShops

Services12,6 12,7 6,5

IndustrialIndustrial buildings

GaragesVacant land

2,4 2,4 4,3

PublicSchools

Public spaces11,4 10,0 5,6

MixedMixed

Land without purpose25,5 28,1 18,1

31

Chapter 6

Conclusion

With the use of 26 volunteers, trips within Sao Paulo have been collected. Only 6 of this vol-unteers made useful trips, with a total of 59 trips being useful within this research. Almost allusers who signed up to participate in the data collection did not make use of the annotationenvironment on an external website and only used the app instead. Also the survey has notbeen filled in by every user. With only 6 out of 26 volunteers responding, a total responserate of 23%, data from the survey could not be used in this research.

Nearly over 60% of all GPS points could be used to calculate the cyclists’ routes, with mostof the excluded points being in a car trip or not being assigned to a trip by the app algorithm.Most of the points and trips are being captured on days with sunny periods with intermediatetemperatures, while on cold cloudy days most points were excluded from the data process-ing. Most points are being collected on weekdays, while most points are being excluded onFridays and Sundays. Most trips are being made on Mondays to Thursdays. Cyclists pre-fer to start their trips during daytime, mostly between 10 and 14 or close to sunrise or sunset.

Most of the speeds in Sao Paulo are within 10 and 30 km/h, with 75% of all road sectionsbeing in this category. Highest speeds are being reached on flat surfaces, with the obser-vation that most of the main routes and cycling facilities are being realised on flat surfaces,possibly causing higher speeds because of the infrastructure instead of the slope. Collectedtrips are nearly 15% longer than the expected route. This might be caused by a preferenceto give up the shortest route to increase other characteristics. But it can also be caused bythe fact that many trips in Sao Paulo are leisure trips, wherein covered distance is irrelevant.More than half of the trips are partly on a cycling facility. Collected and expected routes haveon average the same amount of distance on a ciclofaixa, while collected routes make lessuse of ciclovias then expected. Along the collected and expected routes there is with almost50% mainly a residential land use, however this is less than average compared to Sao Pauloas a whole. Land use that appears more than average along the route, and can therefore beseen as favourable, is mixed (26%), commercial (13%) and public (11%) land use. A quarterof the collected trips have average slopes within 0.5% of the slopes of the expected trips. Athird of the collected trips have a route that is more than 2% hillier than the expected route.

32

Chapter 7

Discussion

Within this chapter the different opportunities for improvement have been explained, as wellas the options for further research as an answer on this document.

Capture areas: Almost all trips that have been generated within this research are within thesouth western area of the centre of Sao Paulo. Within this area a high-level of cycling canbe achieved, since there are a lot of cycling facilities, This can also be seen in Figure 7.1,which shows the percentage of total cycling infrastructure per neighbourhood. In order toguarantee that the results are representative for the whole city of Sao Paulo, more cyclistsare needed and also the dispersion of the cycling trips should be larger.

(a) Ciclovia percentages per neighbourhood (b) Ciclofaixa percentages per neighbourhood

Figure 7.1: Percentage of total cycling infrastructure per neighbourhood

33

Data collection: With the use of the app, detailed routes of the users could be found. How-ever the annotation environment within the app has barely been used during this research.Since the app is using two different ways to collect the data, both from the app itself andfrom the annotation environment in a web browser, the user needs to invest a lot of timein order to take part in the research. Besides taking part in the data collection is entirelyvoluntary, the volunteers need to be given the least amount of nuisance. For future researchit would therefore be useful to merge both environments into one. This could be done byimplementing a feedback phase within the app, which asks people to annotate the trips theyjust made. Another important aspect for future research would be the collection of possi-ble volunteers. Within this research success rate was very high when approaching peoplepersonally, whereas online approach of volunteers turned out to be disappointing. No expla-nation for this could be found, which makes it interesting for future studies.

Identifying car and bicycle trips: Because the GPS tracker of this research was within thesmartphone of the users, all of the trips from the users were captured. Thus a mixture of alltransport modes could have been present in the data processing, with only obvious car tripswith speeds higher than 50 km/h to be filtered out. Being able to filter out only the cyclingtrips would highly benefit the validity of the research. Schussler & Axhausen (2015) comesup with numbers to automatically distinguish the different transport modes with the use ofthe GPS points. With this numbers the differences in transport mode can be found, makingit easier to process the data. No acceleration could be found with the existing data. Howeverfuture data collection could captured and include this data to deal with this issue.

Table 7.1: Automatic cycling trips detection, adapted from Schussler & Axhausen (2015)

Median Speed 95 Percentile Acceleration 95 Percentile SpeedOption 1 0-2 m/s 0.5-1.2 m/s2 -Option 2 0-2 m/s More than 1 m/s2 -Option 3 1.5-6 m/s Less than 0.6 m/s2 0-8 m/s

Follow-up possibilities. Because more information is now available about the most usedroutes, specific solutions to increase the cycling experience can be investigated. It can befurther investigated what the people in Sao Paulo think about their route choices to find theirstated and revealed preference, differences what people tend to do and what they actuallydo. Besides, now that commonly used routes have been found, next steps can be madeto come up with solutions to make these routes more bicycle friendly. This could be doneby implementing traffic calming improvements on the routes, or looking at the differences intraffic flows when certain routes will be shut down entirely for motorised traffic.

34

Bibliography

Andrade, M. F., Kumar, P., de Freitas, E. D., Ynoue, R. Y., Martins, J., Martins, L. D.,Nogueira, T., Perez-Martinez, P., de Miranda, R. M., Albuquerque, T., Goncalves, F. L. T.,Oyama, B. & Zhang, Y. (2017), ‘Air quality in the megacity of Sao Paulo: Evolution overthe last 30 years and future perspectives’, Atmospheric Environment 159, 66–82.

Ben-Akiva, M. & Bierlaire, M. (1999), Discrete choice methods and their applications to shortterm travel decisions, Vol. 23, Springer, Boston Massachusetts.

Bernardi, S., Puello, L. L. P. & Geurs, K. (2017), ‘Evaluation of dutch cycling patterns: Evi-dence from smartphone data’, Journal of Transport and Land Use . Conference version.

Bernardi, S. & Rupi, F. (2015), ‘An analysis of bicycle travel speed and disturbances onoff-street and on-street facilities’, Transportation Research Procedia 5, 82–94.

Bikeplus (2017), ‘Public bike share users survey results 2017’.

Brezina, T., Ibesisch, N., Niegl, M. & Lemmerer, H. (2012), ‘Requirements for high qualitycycling infrastructure design’, International Conference on Road and Rail Infrastructure .Conference version.

Broach, J., Dill, J. & Gliebe, J. (2012), ‘Where do cyclists ride? A route choice model devel-oped with revealed preference GPS data’, Transportation Research Part A 46, 1730–1740.

Buehler, R. & Dill, J. (2016), ‘Bikeway networks: A review of effects on cycling’, Transporta-tion Reviews 36, 9–27.

Centro de Estudos da Metropole (2018), ‘Cycling infrastructure road network’, http://web.fflch.usp.br/centrodametropole/en/.

Chandra, S., Jimenez, J. & Radhakrishnan, R. (2017), ‘Accessibility evaluations for night-time walking and bicycling for low income shift workers’, Journal of Transport Geography64, 97–108.

de Hartog, J. J., Boogaard, H., Nijland, H. & Hoek, G. (2010), ‘Do the health benefits ofcycling outweigh the risks?’, Environmental Health Perspectives 118, 1109–1116.

Dill, J. (2009), ‘Bicycling for transportation and health: The role of infrastructure’, Journal ofPublic Health Policy 30, 95–110.

35

Fishman, E. (2016), ‘Cycling for transport’, Transport Reviews 36, 1–8.

Freitas, A. L. P. & Maciel, A. B. L. (2017), ‘Cycling in a Brazilian city’, Procedia Engineering198, 411–418.

Frejinger, E. (2008), Route Choice Analysis: Data, Models, Algorithms and Applications,PhD thesis, Ecole Polytechnique Federale de Lausanne.

GeoSampa (2018), ‘Mapa digital da cidade de Sao Paulo’, http://geosampa.prefeitura.sp.gov.br/PaginasPublicas/_SBC.aspx#.

Heinen, E., van Wee, B. & Maat, K. (2010), ‘Commuting by bicycle: An overview of theliterature’, Transport Reviews 30, 59–96.

Hood, J., Sall, E. & Charlton, B. (2011), ‘A GPS-based bicycle route choice model for SanFrancisco, California’, The International Journal of Transportation Research 3, 63–75.

Kuiper, D. & Braakman, B. (2009), Five main requirements for cycling-inclusive infrastruc-ture. In Interface for Cycling Expertise, Cycling Inclusive Policy Development: A Hand-book.

Lemas, L. L. & Neto, H. W. (2014), ‘Cycling infrastruture in Sao Paulo: Impacts of a leisure-oriented model’, 12th Annual Conference of the International Association for the Historyof Transport, Traffic and Mobility .

Lowry, M. B., Callister, D., Gresham, M. & Moore, B. (2013), ‘Using bicycle level of serviceto sssess community-wide bikeability’, Trasnportation Research Record: Journal of theTransportation Research Board 2314, 41–48.

MEILI Devs (2018), ‘Pedaladas Beta GPS tracking app’.URL: https://play.google.com/store/apps/details?id=com.airmee.footprints.staging

Menghini, G., Carrasco, N., Schssler, N. & Axhausen, K. (2010), ‘Route choice of cyclists inZurich’, Transportation Research Part A 44, 754–765.

Orenstein, J. (2017), ‘Qual e a diferenca entre ciclovia, ciclofaixa e ciclorrota. e para quaissituacoes elas servem’, Nexo Jornal .

Pritchard, J. P. (2018), Bike network Sao Paulo.

Pucher, J. & Buehler, R. (2008), ‘Making cycling irresistible: Lessons from the Netherlands,Denmark and Germany’, Transport Reviews 28, 495–528.

Ribeiro, F. N. D., de Souza, L. A. T., Salinas, D. T. P., Soares, J., de Oliveira, A. P. & de Miran-dal, R. M. (2015), ‘Air quality in Sao Paulo Brazil temporal evolution and spatial distributionof carbon monoxide coarse particulate matter and ozone’, 9th International Conference onUrban Climate .

36

Rolnik, R. & Klintowitz, D. (2011), ‘(Im)mobility in the city of Sao Paulo’, Estudos Avancados25.

Sa, T. H., Duran, A. C., Tainio, M., Monteiro, C. A. & Woodcock, J. (2016), ‘Cycling in SaoPaulo: Correlates, time trends and health consequences’, Preventive Medicine Reports4, 540–545.

Schussler, N. & Axhausen, K. W. (2015), ‘Processing GPS raw data without additional infor-mation’, Transportation Research Record 2105, 28–36.

Sener, I. N., Eluru, N. & Bhat, C. R. (2009), ‘An analysis of bicycle route choice preferencesin Texas, U.S.’, Transportation 36, 511–539.

Snizek, B., Nielsen, T. A. S. & Skov-Petersen, H. (2013), ‘Mapping bicylists’ experiences inCopenhagen’, Journal of Transport Geography 30, 227–233.

Stinson, M. A. & Bhat, C. R. (2004), ‘A comparison of the route preferences of experiencedand inexperienced bicycle commuters’, Transportation Research Board .

Sustrans (2014), ‘Principles and processes for cycle friendly design (draft)’, Sustrans designmanual .

Trimble Data (2015), ‘Elevation shapefile of Sao Paulo’, https://data.trimble.com/.

Watts, J. & Forner, F. (2017), ‘Resistance! Sao Paulo’s homeless seize the city’, TheGuardian .

Winters, M., Brauer, M., Setton, E. M. & Teschke, K. (2013), ‘Mapping bikeability: a spatialtool to support sustainable travel’, Environment and Planning B: Planning and Design40, 865–883.

Zimmerman, M., Mai, T. & Frejinger, E. (2016), ‘Bike route choice modelling using GPS datawithout choice set of paths’, CIRRELT .

37

Appendix A

Pedaladas app

Within this appendix a manual for the app has been written, as well as the options that canbe annotated in the browser environment.

Manual

The Pedaladas Beta app can be installed on Android via the Google Play Store. After down-loading, an account needs to be made in order to be able to distinguish the different users.Login in with your own credentials will bring you to the home screen, where a switch willmake you able to start collecting data (it will turn red). After turning on the data collection,keep the app running on the background and it will automatically store the GPS points every15-20 seconds and record all your trips to the database.

At the end of every day, all the trips will be available at the annotation environment of the app.This environment can be reached throughout your web browser by going to the followinglink: http://footprints-staging.airmee.com.s3-website-eu-west-1.amazonaws.com/

dash. In the annotation environment the trips will be presented in chronological order, mak-ing you able to correct any mistakes or add notes to the made trips. Options that are availablewithin this environment can be seen in Section A.

Annotation options

Table A.1 gives the different options that can be implemented within the annotation envi-ronment. Characteristics that are or may be already inserted are the travel time, transportmode and destination, depending on whether the app distinguishes the trip to be made inan earlier stage before. All characteristics can be changed, even if the characteristics arepredefined.

38

Table A.1: Pedaladas options

Characteristic Options

Wrong data collection Missed stop between trips or merge with next tripTravel time Start and finishing timeTransfer time Start and finishing time

Transport mode

By footBicycleElectric bicycleCar

TrainBusMetroOther

Facility rating Point on map: rating between 1 and 5 starsDestination Point on mapProblem on the road Point on map

Purpose

Return homeTravel to workTravel to schoolPick-up or drop-off childrenBusiness travelShoppingSport/hobby related travelOther private related travel

Feedback

Rating between 1 and 5 stars and one of the following:This route is the fastestThis route is the calmestThis route is the least pollutedThis route has the best bike pathsThis route has the least amount of trafficThis route is the flattestOther reason, please describe

39

Appendix B

Survey questions

Q1: Thank you for taking part in our research on cycling in Sao Paulo. This research ispart of a collaboration between the University of Twente (UT) in the Netherlands, and thePolytechnic School of the University of Sao Paulo (USP). This questionnaire will take approx-imately 5 minutes to complete. In addition to the questionnaire, our research also makes useof an application that collects data on the way that cyclists use the bicycle in Sao Paulo. Theinstallation of this application is entirely voluntary and all information obtained by it will beanonymised and kept confidential.

Q2: What is your gender?

• Male

• Female

Q3: What is your age?

• Under the age of 18

• 18 - 24 years old

• 25 - 34 years old

• 35 - 44 years old

• 45 - 54 years old

• 55 - 65 years old

• 65 years or older

Q4: What is your occupation?

• Student

• Housewife/house-husband

40

• Employed with a formal contract

• Employed without a formal contract

• Unemployed

• Retired due to disability

• Retired

• Other:

Q5: Do you own a bicycle yourself?

• Yes

• No

Q6: Do you own another mode of transport?

• Yes, a car and a motor

• Yes, only a car

• Yes, only a motor

• No

Q7: How often are you using the bike as a transport mode?

• Never

• Less than 6 times a year

• Between 6 and 11 times a year

• Around three times a month

• One, two or three times a week

• Four or more times a week

Q8: How much experience do you have with cycling in Sao Paulo?

• I started cycling less than a month ago

• I started cycling between a month and half a year ago

• I started cycling between half a year and a year ago

41

• I started cycling between one and two years ago

• I started cycling between two and five years ago

• I started cycling more than five years ago

Q9 What is your main purpose when cycling?

• Work

• Study

• Shopping

• Exercising

• Leisure

• Other:

Q10: What is the e-mail address that you used in the Footprints App?

Q11: What is the importance of the following factors when choosing your route?

Notimportant

A bitimportant

ImportantVery

importantThe mostimportant

Existence of cycling path ◦ ◦ ◦ ◦ ◦Existence of cycling lanes ◦ ◦ ◦ ◦ ◦

Well paved streets ◦ ◦ ◦ ◦ ◦Streets without steep slopes ◦ ◦ ◦ ◦ ◦

Streets without signalcontrolled intersections

◦ ◦ ◦ ◦ ◦

Streets withoutmotorised traffic

◦ ◦ ◦ ◦ ◦

Streets withoutnon-motorised traffic

◦ ◦ ◦ ◦ ◦

42

Q12: What is the importance of the following factors when choosing your route?

Notimportant

A bitimportant

ImportantVery

importantThe mostimportant

The route is the fastest ◦ ◦ ◦ ◦ ◦The route has

the least amount of slopes◦ ◦ ◦ ◦ ◦

This route isthe most attractive

◦ ◦ ◦ ◦ ◦

This route providesthe most security

◦ ◦ ◦ ◦ ◦

This route providesclear directions

◦ ◦ ◦ ◦ ◦

43

Appendix C

Bike Network Sao Paulo

Figure C.1: Cyclable road network of the city of Sao Paulo

44